Automatic recognition of traffic signs is crucial for vehicle safety in autonomous driving. To improve recognition in poor lighting conditions, an enhanced NanoDet-based traffic sign recognition algorithm is proposed. This algorithm introduces an SSM module and a CBAM attention module into the backbone network of the NanoDet model to boost accuracy under adverse lighting. The weighted bidirectional feature pyramid is used to strengthen the feature extraction. To reduce the number of model parameters and increase the viewshed, deep separable convolution is used to replace the standard convolution in the AGM module. The results on the expanded CCTSDB dataset show a speed of 138.2 frames/s and an accuracy of 90.2%, improving 4.7% compared to standard model.